Google Designs Custom Chip for Gemini

Alphabet is reportedly turning its attention toward custom hardware development to refine the operational capacity of its Gemini artificial intelligence suite. As the competition for AI dominance intensifies, the company is looking to move beyond third-party dependencies by creating its own silicon, specifically targeting substantial improvements in power-to-performance metrics.

Project Frozen v2 Development

The internal initiative, currently identified as Frozen v2, is projected for a release window in 2028. According to reports based on anonymous sources, the silicon is being architected with a primary focus on token generation efficiency. Should the hardware meet these internal benchmarks, it would represent a significant shift in how Google manages the computational overhead required to maintain its increasingly complex large language models.

Strategic Shift in Hardware Design

The push for in-house silicon is not unique to Google. Major technology firms are attempting to wean themselves off chipmaker Nvidia, which has maintained a long-standing dominance in the AI hardware sector. This industry-wide trend is driven by two factors: the need for specialized inference processors and the recurring pressures surrounding massive infrastructure expenditures. With competitors like OpenAI recently announcing their own inference processors and Anthropic exploring manufacturing partnerships, the race to own the stack has become a baseline requirement for major AI developers.

Corporate Efficiency and Market Response

Alphabet has faced scrutiny over its significant capital commitments, having previously indicated plans to spend between $180 billion and $190 billion on its AI buildout. Investors have been sensitive to these expenditures, often expressing concerns about the long-term return on investment for such deep-tier infrastructure projects. Following the report on the new chip development, Alphabet saw its stock climb 3% on Monday morning, suggesting that market participants are eager to see tangible efforts toward operational cost-saving measures.

“Our teams are constantly researching and experimenting with new innovations to deliver maximum performance and efficiency for our users and customers. While not every project moves into production, this rigorous exploration is central to our full stack approach. By co-designing our hardware and software from the ground up, we ensure our systems are integrated and highly optimized for real-world workloads.”

— Google, in a statement provided to TechCrunch.

• 6 to 10 times more efficient than existing chips.
• 2028 target release timeframe.
• $180 billion to $190 billion in planned AI expenditures.

Implications for AI Infrastructure

For the broader technology ecosystem, this development signals that the primary bottleneck for AI providers is no longer just software innovation, but the physical reality of electricity and hardware costs. If Google succeeds in producing a chip that significantly lowers the cost per token, it could set a new standard for how firms price and scale AI services. For the end user, this might eventually translate into lower latency or more affordable access to advanced models, provided that these hardware gains are passed down through the service layer rather than simply padding bottom-line margins.